Simulation of Large Scale Neural Models With Event-Driven Connectivity Generation.

Area of Science:

• Computational Neuroscience
• Neuroscience Simulation
• Biophysics

Background:

• Accurate simulation of brain structures is a significant challenge in neuroscience.
• Existing models often focus on improving neuron models or simulation schemes.
• Hodgkin-Huxley (HH) models are continuous, lacking explicit spike events crucial for accurate simulation.

Purpose of the Study:

• To develop a hybrid simulation scheme for efficient and accurate modeling of large-scale neural networks.
• To address the limitations of continuous models in capturing discrete neural events.
• To reduce memory consumption without compromising simulation accuracy.

Main Methods:

• Implemented a hybrid simulation scheme combining time-stepping second-order integration of HH-type neurons with event-driven synaptic current updates.
• Developed a spike detection algorithm to accurately determine spike times in continuous HH models.
• Regenerated outgoing connections at each event to avoid storing connectivity data.

Main Results:

• Significantly reduced memory consumption in neural network simulations.
• Preserved execution time and simulation accuracy, particularly for spike times.
• Demonstrated efficiency using the SiReNe software on a large striatum model (10^6 neurons, 10^8 synapses) under normal and Parkinson's conditions.

Conclusions:

• The proposed hybrid simulation scheme offers an efficient and accurate approach for large-scale neural network simulations.
• The method effectively handles the continuous nature of HH models by incorporating precise spike detection.
• This advancement is crucial for studying complex brain dynamics and neurological conditions.